Reflections and DACs

DonH50

Member Sponsor & WBF Technical Expert
Jun 23, 2010
3,527
3
38
Monument, CO
#1
This article (thread, post, whatever) is to expand upon earlier discussions about reflections and their potential for harm in a digital transmission system as applied to the audio world. I will be qualitative as much as possible, but numbers are bound to creep in… This is also not meant to be a rigorous analysis, more a hand-waving explanation to help folk see what is happening in their system.

First consider a digital driver (source) and load (receiver) connected by a transmission line (cable). The driver and cable are matched with impedance Zo, and the load has mismatched impedance Zx as shown in the figure below. At the top, a single pulse is launched from the source toward the load. The pulse, now a forward signal, travels down the line to the receiver. Even at nearly the speed of light it takes a little time to get there.

Fig 1..JPG

As the signal hits the load, the difference in impedances causes part of it to “bounce back” or reflect from the load. How much depends upon the mismatch; no mismatch, no reverse wave. The reflected signal, which is actually inverted from the original, travels back toward the source (the reverse signal).

In this example, the reverse signal meets the next pulse from the driver near the source end of the line. Where that actually happens depends upon the pulse (bit) rate and length of cable. When the forward and reverse waves meet, some change in the forward signal occurs as the reverse signal interacts with it. In this case, destructive summation is shown, as the out-of-phase reverse signal cancels some of the forward signal pulse. If you hold a hose that has water flowing out of it in your left hand, and try to mate it to a different hose in your right hand, what usually happens is you don’t get perfect alignment and some of the water squirts back on your left hand. Same sort of idea…

Now, Figure 1 shows a periodic pulse train that might be produced by a string of digital data that simply alternates (0,1,0,10,1…) every 250 ns (4 Mb/s). Red is the original signal from the source, blue shows the reflected wave (1/10th amplitude and 180 degrees out of phase), and green shows the resulting pulse train. I have deliberately placed the edges of the reflected wave coincident with the edges of the forward (source) wave.

Fig 2..jpg
Figure 1. Periodic pulse train.​

Now, reducing the amplitude has to change the edge (slew) rate, so this could cause the actual center (zero) crossings to move. That would cause jitter. However, as Figure 2 shows, for an ideal periodic pulse sequence like this, no damage is done. This shows the period measured (histogram) for each of 100 edges. The red box represents the ideal signal and is a single period; the box’s width merely makes it easier to see. The blue stem line represents the signal at the load after reflection. If any period was not exactly 250 ns we would see additional spikes in the histogram. In this case, both ideal and load signals have 100 “hits” at 250 ns, with no errors. As I am tired of making pictures, you’ll have to trust me when I say changing the phase of the reverse wave (moving it with respect to the forward wave) does not matter – a perfect histogram still results.

Fig 3..JPG
Figure 2. Period histogram of crossings for ideal (red) and load (blue) signals.​

While there are many things that can cause variation in even a regular pulse train, let’s move on and consider what happens when the forward and reverse signals are different. After all, a real digital bit stream is not a regular sequence, but varies with the signal. There may be random-length strings of 1’s and 0’s distributed among the signal stream, so in general the signal is not at all regular and periodic like the first example.

Figure 3 shows a reverse signal (blue) three times the length of the source (red). To simplify the simulation, I made it a regular sequence as well, but of course in real life both streams will look essentially random (a function of the signal). In this example, the reverse wave interacts with the bit stream along the cable. We’ll look at the period assuming the bit stream alternates between a stream of single 0/1 transitions (0,1,0,1…) and a 000/111 sequence (0,0,0,1,1,1,0,0,0,1,1,1,…) Notice the output signal (green) is now modulated with the reverse signal. With real data, the forward and reverse signals will in general be different. We again care about what happens at the load, where the two streams interact and the resultant signal is measured by the receiver.

Fig 4..JPG
Figure 3. Reverse signal with period 3x the forward signal​

Figure 4 shows the resulting histogram. Now, the ideal source still has only one bin centered at 250 ns as expected. However, the load signal is now distributed into three bins, indicating three slightly different periods. This is deterministic jitter – that is, jitter related to the signal (and clock) period, length of cable, amount of mismatch, etc.

Fig 5..JPG
Figure 4. Period histogram with different forward and reverse signals.​

So long as the jitter is fairly small, the receiver will have no problem capturing the right data bits. With a 250 ns bit period, no practical receiver would have issues getting the bits right with just a few ns of jitter as shown in this example.

The problem comes when those edges are used to generate the clock for (e.g.) a DAC. The clock circuit will tend to reject small random movements (random jitter), but when patterns like this occur due to a combination of signal bits, clock rate, impedance mismatches, and cable length (among other things), the clock circuit may actually follow the periodic variations and change the clock period accordingly. It has no way of knowing the deterministic jitter is not clock drift, and we get jitter in the output of our DAC. Bummer!

Of course, in the real world there will many different periods and thus many bins of various magnitudes. The jitter threads delve into the magnitude of jitter that might cause problems. In this case, the magnitude is a function of the mismatches among source, load, and cable in the link. Clearly, or perhaps not so clearly, if the cable is the right multiple of the clock frequency (bit period), then any “glitches” will happen at the same time and again no harm done. In general, that is very difficult, as there are a myriad of parameters to control – many of which are in fact beyond our control – and of course the critical length will be different for each data (bit) rate.

Please do not take this as a rigorous proof, but hopefully it helps show some of the things that can go wrong when digital and analog meet, even if we think it’s all digital… - Don
 
Last edited:

DonH50

Member Sponsor & WBF Technical Expert
Jun 23, 2010
3,527
3
38
Monument, CO
#4
Thanks guys!

If who claims rights? Actually, I retain rights (checked that out early on) with WBF. None of this is new or novel, save the way I present it (we are all unique, for better or worse ;) ).

OT: I have been thinking I should make a post with links to the various articles in order (more or less). A summary post, sort of, to help people (OK, me) keep track and find the threads. It would also make my life easier by providing a single focal point instead of searching through all the threads. There are few posts I know I have made that I'd like to quote or send people to wen they ask something, but durn if I can find them! Senility...
 

fas42

Addicted To Best
Jan 8, 2011
3,973
0
0
NSW Australia
#6
Don, just trying to get a handle on your explanation there, with regard to the destructive summation at the source end. The forward and reverse waves anywhere along the line except at the ends just pass through each other without alteration, I think, but I don't quite understand the process at the source end. Could you fill in a bit more there please; and does it vary per the source impedance?

Frank
 

DonH50

Member Sponsor & WBF Technical Expert
Jun 23, 2010
3,527
3
38
Monument, CO
#9
Don, just trying to get a handle on your explanation there, with regard to the destructive summation at the source end. The forward and reverse waves anywhere along the line except at the ends just pass through each other without alteration, I think, but I don't quite understand the process at the source end. Could you fill in a bit more there please; and does it vary per the source impedance?

Frank
Yes, they pass through on the way, but at the ends the signals are terminated (they no longer travel, if you will). The problem is at the receiving end, where reflections modulate the incoming pulses and that is right at the sampling point.

If the termination is perfect, as was assumed at the source, then there are no more reflections from that end. If there is a mismatch, then a small part of the reverse wave would be reflected back again toward the load, and so forth and so on with signals bouncing back and forth. It can get real messy real quick...
 

fas42

Addicted To Best
Jan 8, 2011
3,973
0
0
NSW Australia
#10
Sorry about this, Don, still not making sense for me! Assuming the source is matched correctly, the reflected wave is dissipated in the source impedance. I can see that destructive summation will occur dynamically along the line as the forward and reverse wave travel past each other. But at the load it appears to me that the forward and reverse wave are the one and the same, they are coincident in time and space, there is no separate forward and reverse wave to interact at the load termination.

Hope you can shed more light ...

Frank
 

DonH50

Member Sponsor & WBF Technical Expert
Jun 23, 2010
3,527
3
38
Monument, CO
#11
OK, I see your point. Part of this is because I simplified greatly and so the example lacks rigor. At the load point there is only one voltage (or current) at any instant in time, and it is formed by the summation of all the forward and reverse waves bouncing back and forth along the cable at that point (and time). So your assumption is true, but does not take into account all the other waves that occur in a real system. The various waves do not all arrive at the same time and with the same amplitude or polarity. The result of all these waves bouncing around is that the load voltage varies with the signal (among other things) if both source and load are not perfectly matched.

I did not show this in my simple example and took liberties to provide a hand-waving version understandable to most (I hope). It will not hold up to in-depth analysis, but I did not want to try to cram even an introductory course in transmission line theory into it.

HTH -- gotta' go practice! - Don
 

Orb

New Member
Sep 8, 2010
3,022
0
0
#13
Vincent quick comment about the table on your page that you may want to check - nearly spot on but bits wrong.
It should be 32 bits not 24 as 32bits is the size of the S/PDIF AES frame.
This then gives us a bit rate frequency of 2.8mhz.

Sorry got to go now but check it out, should come out to that but I appreciate I may be wrong.

Cheers
Orb
 

DonH50

Member Sponsor & WBF Technical Expert
Jun 23, 2010
3,527
3
38
Monument, CO
#16
BNC s are better than most RCAs, but are still not a great RF connection. Look up SMA and the like. Of course, manufacturers won't add expensive connectors without some sort of market push... I have seen some matched RCAs but it's been a while and I have no idea if they are still available. I should do a little research and see what the impedance of RCAs really is. Maybe I can piddle with the new VNA we got at work... :)

Here's a scary thing: some of the very high frequency connectors we use cost $100 to $300 or more just for the connector! Be glad HDMI doesn't extend into the 50+ GHz region...
 

1audio

New Member
Sep 17, 2010
29
0
0
#17
A well made 75 ohm BNC can measure really well up to 3 GHz on a VNA. The Smith chart is a dot in the middle as long as its all well cared for. You would need the BNC adaptor Kit from HP. $1,000 for some 75 Ohm BNC to N adapters. The test cable for 75 Ohm N to N for Agilent VNA's is around the same price. Suddenly audiophile cables seem less ridiculous, however none will measure nearly as well on a VNA.

I have measured a number of RCA connectors with a TDR and none were higher than about 35 Ohms. Its not possible without something like a "wormhole' to reverse time. The ID to OD ratio of an RCA prevents anything else. However some argue that the length of an RCA is so short it won't have an effect on the data. The AES3 standard calls out a rise time of 5nS to 30 nS which would seem to make a 2 inch disturbance from an RCA not important. I don't buy that. And BNC connectors work much better than RCA's, very rarely tearing the connector from the chassis when trying to remove them.

Making HDMI extend to 20+ GHz (40 Gbps) is just around the corner. Its all about the biggest number, you know. (The fastest receiver chip available today stops at 6.68 Gbps for three lanes but will need to be supplanted for 4K displays so that will happen in the next 12 months.)
 

DonH50

Member Sponsor & WBF Technical Expert
Jun 23, 2010
3,527
3
38
Monument, CO
#18
Thanks Demian. I have little experience with 75-ohm BNCs; my world has been mostly 50 ohms. HP (sorry, Agilent now) uses a modified BNC that is good to 10+ GHz on their high-end real-time DSOs. Needless to say (so I will anyway), it is not a run-of-mill BNC in performance or price.

I have measured a skeletonized RCA with a special pin some company made that was higher -- don't recall if it made 75 ohms.

I have not been following HDMI; are they planning to put 40 Gbps on a single lane? That's a little scary... Unless the receivers tolerate a lot of loss (and/or have great ERC) cables are going to be costly. SAS is struggling with 6 Gb/s in some systems, let alone the upcoming 12 Gb/s lanes. Not sure I have seen 40 Gb/s over wire, but I admit I have not kept up on networking (did work on a 20G VCSEL driver last year).

Though not really clear from my simple explanation in this thread, edge speed is a crucial parameter when considering reflections and how long (or short) a discontinuity matters. As Demian says, the inch that does not matter at a few MHz with slow edges, becomes critical at high data rates.
 
Last edited:

ar-t

New Member
Jun 3, 2011
73
0
0
Texas
ar-t.co
#19
I'll second the "35 ohms max" measurement. As for the "skeletonized" RCAs, the ones they claim are 75 ohms.........

What happens when you mate them, with a conventional RCA? You are right back where you started. So, if you are going to use a special connector, 75R BNCs are not hard to find.
 

DonH50

Member Sponsor & WBF Technical Expert
Jun 23, 2010
3,527
3
38
Monument, CO
#20
Not sure why the argument? I think we all agree on the need for good connectors... I do not recall how high the skeletonized RCAs got but I am virtually certain it was not 75 ohms. I would not have been looking for that anyway; my world is 50 ohms (RF/mW/mmW), and we don't use RCAs for numerous reasons (lousy impedance, low bandwidth, and no positive capture being among them). Cutting one up was for a lab test, not to be used in the real world. After all these years, I am not even sure why we did it...

I'll second the "35 ohms max" measurement. As for the "skeletonized" RCAs, the ones they claim are 75 ohms.........

What happens when you mate them, with a conventional RCA? You are right back where you started. So, if you are going to use a special connector, 75R BNCs are not hard to find.
 

About us

  • Founded in 2010 What's Best Forum invites intelligent and courteous people of all interests and backgrounds to describe and discuss the best of everything. From beginners to life-long hobbyists to industry professionals we enjoy learning about new things and meeting new people and participating in spirited debates.

Quick Navigation

User Menu

Steve Williams
Site Founder | Site Owner | Administrator
Ron Resnick
Site Co-Owner | Administrator
Julian (The Fixer)
Website Build | Marketing Managersing